It's a bit awkward, but I trained this version on just 13 images. Nevertheless, I genuinely tried to minimize the LoRA's tendency to fixate on repetitive elements. I made an effort to eliminate checkered patterns, collars, ambiguous eyes, and many other things that made me cringe. Nevertheless, the LoRA is capable of delivering high-quality and pleasing results.

Shenki Style v8 (NSFW)

Instead of 32 images, I kept only 13 worthy ones. Unlike the previous v7, this one uses 32 alpha and 64 dim.

Files with SHA256: d9a4b0dcaa9d21ad8182e21ae173de80a6cd6d134627545356217ab65fa96920